A remote sensing data parallel processing method for soil water content inversion

Abstract

This invention relates to the field of parallel data processing technology, specifically a parallel processing method for remote sensing data for soil moisture content retrieval. The method includes the following steps: acquiring common parameters from multi-source images; determining the number of threads and segmenting pixels based on hardware resources; constructing parallel tasks and address mappings to generate isolated copies; inverting rough surface scattering to obtain the scattering cross-section of bare soil; calculating vegetation attenuation and filtering out vegetation scattering; and generating parallel results for soil moisture content retrieval. In this invention, through parallelization, processing time is significantly reduced and CPU utilization is improved, solving the timeliness bottleneck of the original algorithm. Simultaneously, the parallel calculation results are completely consistent with the serial results without sacrificing accuracy. Its architecture supports horizontal scaling, can adapt to more CPU cores, and is suitable for larger-scale data processing scenarios. By optimizing resource scheduling and task distribution mechanisms, it provides efficient technical support for scenarios such as agricultural irrigation and drought monitoring.

Description

Technical Field

[0001] This invention relates to the field of parallel data processing technology, and in particular to a parallel processing method for remote sensing data for soil moisture content retrieval. Background Technology

[0002] Data parallel processing technology refers to a branch of computer science that breaks down complex computational tasks and distributes them to multiple computing units for synchronous execution. Its related content involves distributed computing architectures and multi-task scheduling mechanisms. Its main purpose is to reduce the processing cycle of massive amounts of information through the allocation of computing resources. This technology is widely used in industries involving large-scale data processing needs, such as weather forecasting, geological exploration, and financial analysis. Traditional remote sensing data parallel processing methods refer to a computational model that divides large-scale Earth observation images into several sub-tasks and distributes them to different nodes for synchronous processing. This method is mainly used to analyze large and complex satellite observation images. In practical applications, existing technologies such as cluster servers, distributed file systems, and information transmission interfaces are typically used for data segmentation and task communication allocation.

[0003] Existing technologies primarily rely on water cloud models, advanced integral equation models, and Dobson models for soil moisture inversion. However, the serial computing architecture leads to severe bottlenecks when processing large-scale remote sensing data: high computational complexity and numerous iterations result in exponentially increasing time overhead and low utilization of CPU multi-core resources, failing to meet the timeliness requirements of actual business operations (such as real-time environmental monitoring or disaster response). The root cause of this problem lies in the fact that traditional methods do not fully utilize the parallel computing capabilities of modern multi-core CPUs, leading to wasted hardware resources and system response delays. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the above-mentioned traditional technologies and provide a parallel processing method for remote sensing data for soil moisture content inversion.

[0005] The objective of this invention is achieved through the following technical measures: A parallel processing method for remote sensing data for soil moisture content retrieval includes the following steps: S1: Acquire multispectral remote sensing images and microwave radar remote sensing images; extract vegetation indices based on band reflectance in the multispectral remote sensing images; map incident angle parameters according to observation data in the microwave radar remote sensing images; combine the vegetation indices with the incident angle parameters to generate preprocessing common parameters. S2: Obtain the total number of microwave image pixels and microwave image data storage capacity from the microwave radar remote sensing image, extract the cache capacity threshold of the local computing node, determine the number of parallel threads in combination with the number of available hardware cores, and perform spatial traversal and segmentation of the raster space corresponding to the total number of microwave image pixels according to the number of parallel threads to obtain independent pixel copies. S3: Construct a parallel execution task that matches the number of parallel threads, distribute the independent image data copy and the preprocessing common parameters to the thread memory address space allocated by the parallel execution task, perform access boundary locking and dedicated memory address mapping on the independent image data copy, and generate an isolated image data copy; S4: From the isolated image data copy, map the radar microwave roughness surface scattering correlation between the surface roughness parameters and the incident angle conversion parameters to generate roughness surface scattering features, and invert the underlying microwave reflection parameters between the roughness surface scattering features and the surface dielectric constant data to obtain bare soil scattering cross section data. S5: Map the vegetation canopy radar microwave transmittance features between the vegetation index data and the incident angle conversion parameters to generate vegetation radar transmittance parameters; extract the vegetation canopy microwave attenuation ratio between the vegetation radar transmittance parameters and the bare soil scattering cross section data to generate vegetation layer attenuation; filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated pixel replica; and generate parallel processing results for soil moisture content inversion.

[0006] As a further aspect of the present invention, the red band reflectance and near-infrared band reflectance of multispectral pixels are extracted from the multispectral remote sensing image, and the vegetation index data are generated by band feature fusion extraction based on the red band reflectance and near-infrared band reflectance of multispectral pixels. Satellite radar observation incident angle data is extracted from the microwave radar remote sensing image, and the units of the satellite radar observation incident angle data are unified from angle to radians to generate the incident angle conversion parameter.

[0007] As a further aspect of the present invention, if the microwave image data storage capacity is greater than the cache capacity threshold, a first thread allocation instruction is generated based on the mapping relationship between the number of available hardware cores and the number of cores. If the microwave image data storage capacity is not greater than the cache capacity threshold, then a second thread allocation instruction is generated based on the equal number of available hardware cores. The target concurrency level is determined based on the first thread allocation instruction or the second thread allocation instruction, and the number of parallel threads is generated.

[0008] As a further aspect of the present invention, the roughness surface scattering characteristic specifically refers to the radar microwave roughness surface scattering correlation between the surface roughness parameter and the incident angle conversion parameter.

[0009] As a further aspect of the present invention, step S1 specifically comprises: S111: Acquire multispectral remote sensing images and microwave radar remote sensing images, extract multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance from the multispectral remote sensing images, perform band feature fusion extraction on multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance, and generate vegetation index data. S112: Based on microwave radar remote sensing imagery, the observation data is mapped to extract satellite radar observation incident angle data. The satellite radar observation incident angle data is then converted to radians. The processed data is then converted and processed in conjunction with the microwave radar remote sensing imagery observation data mapping relationship to obtain the incident angle conversion parameters. S113: Perform sequence combination and splicing processing based on the vegetation index data and the incident angle conversion parameters. Perform spatial correspondence set combination on the corresponding dimension data columns of the vegetation index data and the incident angle conversion parameters. List multi-dimensional parameters according to the combined data items and extract the corresponding observation data items to construct a unified format architecture and establish preprocessing common parameters.

[0010] As a further aspect of the present invention, step S2 specifically includes: S211: Acquire microwave radar remote sensing images, extract the total number of microwave image pixels and microwave image data storage capacity, collect the number of available hardware cores of local computing nodes, simultaneously extract the local computing node cache capacity threshold, summarize various hardware-related parameters and image storage parameter characteristics, and generate node hardware basic characteristic data. S212: Based on the microwave image data storage capacity in the node hardware basic feature data, compare it with the cache capacity threshold. If the microwave image data storage capacity is greater than the cache capacity threshold, perform a double core quantity relationship mapping in combination with the number of available hardware cores. If it is not greater, perform an equal core quantity relationship mapping to obtain thread allocation instruction parameters. S213: Based on the thread allocation instruction parameters, determine the target number of concurrent operations and convert it into a parallel thread numerical index. According to the converted parallel thread numerical index, perform spatial traversal and segmentation on the grid space corresponding to the total number of microwave image pixels, extract the set of associated pixels of each grid image, and establish an independent pixel copy.

[0011] As a further aspect of the present invention, step S3 specifically comprises: S311: Create parallel environment parameters for the independent cell replica, construct a matching parallel execution task set in combination with the parallel thread numerical index, allocate corresponding thread computing resources and memory resource quotas to each independent execution unit within the parallel execution task set, and perform task packaging and allocation processing to obtain parallel execution task data. S312: Based on the parallel execution task data, obtain and allocate the thread memory address space, distribute the independent cell copy and the previously mentioned preprocessing common parameters to the thread memory address space for storage management and synchronization operations, establish mutually independent parallel computing storage segment space, and generate memory address distribution parameters; S313: Based on the memory address distribution parameters, perform access boundary locking operations and out-of-bounds read / write interference shielding on the independent cell replicas within the thread memory address space. Combined with the access boundary locking operation, perform dedicated memory address mapping operations on memory segments to separate concurrent data items of each thread and establish isolated cell data replicas.

[0012] As a further aspect of the present invention, step S4 specifically comprises: S411: Based on the isolated image data copy, perform parameter deconstruction, extract the surface radar backscattering coefficient and surface roughness parameters, perform matrix serialization and rearrangement of the extracted surface radar backscattering coefficient and surface roughness parameters, summarize the obtained data sequence and establish a radar microwave parameter set to obtain the surface microwave basic parameters. S412: Extract the incident angle conversion parameter from the preprocessed common parameters, and perform radar microwave roughness surface scattering mapping on the correlation between the surface roughness parameter and the incident angle conversion parameter in the surface microwave basic parameters using an advanced integral equation model to generate roughness surface scattering features. S413: The roughness surface scattering characteristics are imported into the advanced integral equation model. The underlying microwave reflection parameters between the roughness surface scattering characteristics and the surface dielectric constant are inversely inverted. The microwave reflection parameter terms obtained from the inverse inversion operation are integrated to construct a microwave inversion cross-sectional data sequence, and the bare soil scattering cross-sectional data are output.

[0013] As a further aspect of the present invention, step S5 specifically comprises: S511: Extract vegetation index data and incident angle conversion parameters according to the preprocessing common parameters, use the water cloud model to perform a positive mapping of the vegetation canopy radar microwave transmittance characteristics between the vegetation index data and the incident angle conversion parameters, construct a transmittance representation vector, and obtain the vegetation radar transmittance parameters. S512: Input the vegetation radar transmittance parameters and the bare soil scattering cross section data into the water cloud model, extract the vegetation canopy microwave attenuation ratio index between the vegetation radar transmittance parameters and the bare soil scattering cross section data, summarize the vegetation canopy microwave attenuation ratio index and quantify the attenuation amplitude corresponding to each calculation node, and generate the vegetation layer attenuation amount. S513: Filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated image data copy using the vegetation layer attenuation amount, perform spatial variable reverse analysis calculation on the residual data, and aggregate the surface radar backscattering cross section inversion analysis output items corresponding to each parallel computing node to obtain the parallel processing result of soil moisture content inversion.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Significantly improves computational efficiency: Through parallelization, processing time is greatly reduced, and CPU utilization is improved, solving the timeliness bottleneck of the original algorithm while maintaining computational accuracy and scalability: Parallel computing results are completely consistent with serial results, proving that the solution improves efficiency without sacrificing accuracy; its architecture supports horizontal scaling, can adapt to more CPU cores, and is suitable for larger-scale data processing scenarios, with strong engineering practicality: The solution has been successfully applied to the soil moisture analysis project in Weifang City, providing efficient technical support for scenarios such as agricultural irrigation and drought monitoring by optimizing resource scheduling and task distribution mechanisms, combining theoretical innovation and practical value. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the steps of the present invention.

[0016] Figure 2 This is a flowchart of step S1 of the present invention.

[0017] Figure 3 This is a flowchart of step S2 of the present invention.

[0018] Figure 4 This is a flowchart of step S3 of the present invention.

[0019] Figure 5 This is a flowchart of step S4 of the present invention.

[0020] Figure 6 This is a flowchart of step S5 of the present invention. Detailed Implementation

[0021] Example: This example provides a parallel processing method for remote sensing data for soil moisture content retrieval. In practical applications, such as during the continuous operation of remote sensing inversion of farmland surface conditions, this method runs at a local processing node, receiving multispectral remote sensing images and microwave radar remote sensing images corresponding to the same area. It performs spatial correspondence, thread isolation, radar scattering analysis, and vegetation scattering filtering on the multi-source images to form a parallel processing result for soil moisture content retrieval. The method includes the following steps: Please see Figure 1 and Figure 2 S1: Acquire multispectral remote sensing images and microwave radar remote sensing images. Extract vegetation indices based on the band reflectance in the multispectral remote sensing images. Map the incident angle parameters according to the observation data in the microwave radar remote sensing images. Combine the vegetation indices and incident angle parameters to generate preprocessing common parameters.

[0022] The preprocessing common parameters are data objects repeatedly read in subsequent parallel processing stages. They consist of vegetation index data, incident angle transformation parameters, and their spatial correspondences. Vegetation index data is extracted from multispectral remote sensing imagery and represents the surface vegetation cover status. The incident angle transformation parameters are data items formed by uniformly expressing the incident angle observed in microwave radar remote sensing imagery. The spatial correspondences ensure that the same surface location can be identified and retrieved by the same pixel processing process in both multispectral and microwave radar data. These preprocessing common parameters only serve a cross-thread shared reading function and do not involve rewriting the current pixel copy. Subsequent steps retrieve the corresponding data items according to the spatial location index when reading these parameters.

[0023] If multispectral or microwave radar remote sensing images have missing spatial references, mismatched pixel locations, or unresolved image metadata, the corresponding images are first marked as pending verification. The missing category, source image type, and affected area are then saved in the preprocessing record. Images with complete spatial correspondence continue to subsequent processing; data items without complete spatial correspondence are not included in the common parameter stitching, and their corresponding spatial locations are output as supplementary markers to prevent subsequent threads from reading data without a source.

[0024] S111: Acquire multispectral remote sensing images and microwave radar remote sensing images, extract multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance from the multispectral remote sensing images, perform band feature fusion extraction on the multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance, and generate vegetation index data.

[0025] Multispectral pixel red band reflectance is pixel reflectance information obtained by resolving the corresponding red channel in multispectral remote sensing images, while multispectral pixel near-infrared band reflectance is pixel reflectance information obtained by resolving the near-infrared channel. Both types of reflectance information are entered into the same band feature fusion process using pixel spatial location as an index. Band feature fusion extraction does not introduce numerical examples; instead, it expresses the relative relationship between the two types of reflectance information at the same spatial location according to engineering rules governing the differences in red light absorption and near-infrared reflectance by vegetation, resulting in vegetation index data representing vegetation cover status.

[0026] During the extraction process, the band identifiers, spatial references, and pixel validity identifiers of the multispectral image are first read. Then, data items with consistent spatial locations in the red band and near-infrared band are paired. If any band has an invalid identifier, duplicate record, or spatial misalignment, a band pairing anomaly identifier is generated for that pixel and written into the preprocessing record. This anomaly location will not be included in the vegetation index data subsequently. Through this processing, the vegetation index data can maintain a clear source, clear spatial location, and clear boundaries for subsequent readings.

[0027] S112: Based on microwave radar remote sensing imagery, the observation data is mapped to extract satellite radar observation incident angle data. The satellite radar observation incident angle data is processed to unify the angle to radian units. The processed data is then combined with the microwave radar remote sensing imagery observation data mapping correlation to perform conversion processing to obtain the incident angle conversion parameters.

[0028] Satellite radar observation incident angle data are metadata from microwave radar remote sensing images or observational geometric data provided synchronously with the images, reflecting the geometric relationship between the radar beam and the observed location on the ground surface. Unified processing from angle to radians ensures that subsequent roughness surface scattering mapping and vegetation canopy transmittance mapping use the same angular expression. The incident angle conversion parameter is a parameter object that has undergone unit unification and is established to correspond with the spatial location of microwave image pixels; its output is used for preprocessing common parameters and subsequent radar scattering processing stages.

[0029] When the incident angle record in a microwave radar remote sensing image cannot directly correspond to the pixel location, spatial matching is first performed based on the mapping relationship between the image observation data. The matching criteria include the image row and column positions, geographic positioning relationships, and observation geometric records. If the incident angle record is missing or there are multiple source conflicts at the same location, an observation angle anomaly marker is generated for the conflict location. Subsequent roughness surface scattering processing does not read the incident angle transformation parameters at that location, and the anomaly marker is passed to the final result record to indicate that no effective inversion output was formed at that location.

[0030] S113: Perform sequence combination and splicing processing based on vegetation index data and incident angle transformation parameters. Perform spatial correspondence set combination on the corresponding dimension data columns of vegetation index data and incident angle transformation parameters. List multi-dimensional parameters according to the combined data items and extract the corresponding observation data items to construct a unified format architecture and establish preprocessing common parameters.

[0031] Sequence combination and stitching processing merges vegetation index data and incident angle transformation parameters into a shareable, readable data structure using the same spatial index. The multi-dimensional parameter list is not a numerical list, but rather a data organization method that records parameter types, source image categories, spatial indices, validity status, and subsequent readable locations. A unified format architecture allows subsequent threads to obtain vegetation index data and incident angle transformation parameters through the same read path, eliminating the need for repeated parsing of the original image during parallel processing.

[0032] When establishing preprocessing common parameters, it is first confirmed that both vegetation index data and incident angle transformation parameters have valid identifiers for the same spatial location. Then, both types of data items are written into the common parameter section. If only one type of data item is valid, that spatial location is marked as incomplete in the common parameters, and this incomplete state is saved along with the common parameters. This state serves as the basis for reading and judging during subsequent processing of isolated image data copies, preventing the absence of common parameters from causing radar scattering mapping or vegetation attenuation processing to have no source input.

[0033] Please see Figure 1 and Figure 3 S2: Obtain the total number of microwave image pixels and the data storage occupancy information of microwave image remote sensing images from microwave radar remote sensing images, extract the high-speed cache capacity determination boundary of local processing nodes, determine the parallel thread configuration in combination with the available hardware core situation, and perform spatial traversal and segmentation of the raster space corresponding to the total number of microwave image pixels according to the parallel thread configuration to obtain independent pixel copies.

[0034] The total number of microwave image pixels describes the spatial coverage of the raster to be processed, while the microwave image data storage occupancy information describes the storage resource status of the microwave radar remote sensing image in the local processing node. The cache capacity determination boundary is a preset resource rule used by the local processing node to determine whether image data is suitable for direct cache-friendly thread processing; it originates from the hardware resource configuration records of the local processing node. The parallel thread configuration is a thread allocation result generated based on the image storage occupancy status and the availability of hardware cores, used to limit the creation scale of subsequent parallel execution tasks and data partitioning boundaries.

[0035] An independent cell copy is a data object obtained by segmenting the corresponding raster space from a microwave radar remote sensing image. It carries a set of non-overlapping cell locations, corresponding microwave radar backscattering information, reference locations for surface roughness parameters, and spatial index identifiers. This copy is only accessed by the corresponding thread and will not be directly overwritten by other threads subsequently. If duplicate, missing, or overlapping spatial indices occur during raster space segmentation, the segmentation records are first rolled back to a stable state, and then the copy boundary records are regenerated according to the spatial traversal order to ensure that the data copies obtained in subsequent steps have independent access boundaries.

[0036] S211: Acquire microwave radar remote sensing images, extract total microwave image pixel information and microwave image data storage and occupancy information, collect information on available hardware cores of local processing nodes, synchronously extract the high-speed cache capacity determination boundary of local processing nodes, summarize various hardware-related parameters and image storage parameter features, and generate basic hardware feature data of nodes.

[0037] The node hardware basic characteristic data is a resource description data object used for thread allocation and raster segmentation. It includes information on the total number of microwave image pixels, image storage occupancy status, available hardware core status, and cache capacity determination boundaries. This data object is read and saved by the local processing node during the task startup phase. After entering the subsequent thread allocation phase, it is only used as a basis for determination and does not participate in the physical parameter mapping of the soil moisture content inversion itself.

[0038] When generating basic hardware feature data for a node, the raster range and data storage occupancy record of the microwave radar remote sensing image are read first. Then, the core hardware status that the local processing node can currently process is read, and the hardware resource status and image storage occupancy status are written to the same resource record. If the hardware resource record is inaccessible or the cache carrying boundary cannot be read, the thread configuration status is marked as pending confirmation, and an executable thread configuration record is formed using the stable configuration already registered by the local processing node. At the same time, this status is written to the task record for subsequent anomaly tracing.

[0039] S212: Based on the microwave image data storage occupancy information within the node hardware basic feature data, compare it with the cache capacity judgment boundary. If the microwave image data storage occupancy information exceeds the cache capacity judgment boundary, then perform extended core mapping in combination with the available hardware core situation. If it does not exceed the boundary, then perform equal core mapping and obtain thread allocation instruction parameters.

[0040] Thread allocation instruction parameters are control data output from the resource determination process, used to indicate the thread concurrency configuration that should be adopted for subsequent parallel execution tasks. Extended core mapping indicates that when image data storage exceeds cache capacity, threads are configured according to a finer-grained concurrency processing direction, keeping the data blocks handled by a single thread within controllable read boundaries. Equal-size core mapping indicates that when image data storage capacity matches cache capacity, thread configurations are generated according to the available hardware core state, maintaining a stable correspondence between threads and hardware resources.

[0041] The comparison process is based solely on resource status categories and preset resource rules, without outputting specific resource values. If the relationship between image storage occupancy status and cache capacity determination boundaries cannot be confirmed, the thread allocation instruction parameters are marked as conservative scheduling, and an allocation direction that does not cause data boundary intersections between threads is selected. These instruction parameters are then read and converted into parallel thread configurations in subsequent steps, serving as the basis for raster space partitioning.

[0042] S213: Based on the thread allocation instruction parameters, determine the target concurrency state and convert it into parallel thread numerical indicators. According to the converted parallel thread numerical indicators, perform spatial traversal and segmentation on the raster space corresponding to the total number of microwave image pixels, extract the set of associated pixels of each raster image, and establish independent pixel copies.

[0043] In this embodiment, the parallel thread numerical index only indicates the thread configuration status and does not introduce a specific number of threads. Spatial traversal segmentation is performed according to the continuous access order of the microwave image raster space, so that each independent pixel copy corresponds to a defined spatial region or a set of continuously traceable pixel positions. The raster image associated pixel set contains pixel spatial index, microwave radar backscatter data reference, surface roughness parameter reference, and validity identifier. Subsequent steps use this set as the basis to establish thread-isolated data copies.

[0044] If invalid pixels, holes, or inconsistent boundary records exist in the raster space, the segmentation process will retain the corresponding locations as anomalous spatial identifiers and write these identifiers into the metadata area of ​​the independent pixel copy. These locations will not be directly used as valid inversion objects in the scattering process, but their spatial identifiers will be retained in the result stitching stage to maintain consistency between the output results and the spatial extent of the original microwave radar remote sensing image.

[0045] Please see Figure 1 and Figure 4 S3: Construct parallel execution tasks that match the parallel thread configuration, distribute independent image metadata copies and preprocessing common parameters to the thread memory address space allocated by the parallel execution tasks, perform access boundary locking and dedicated memory address mapping on the independent image metadata copies, and generate isolated image metadata copies.

[0046] Parallel execution tasks are executable processes established by local processing nodes for independent cell replicas, inheriting the thread configurations and independent cell replicas output by S2. The thread memory address space is an independent read-write area allocated for the corresponding parallel execution task, used to store the independent cell replicas accessible to the current thread, references to common parameters, and processing status records. Access boundary locking restricts threads to reading and modifying cell data in their own replicas only, while dedicated memory address mapping establishes a correspondence between each independent cell replica and a unique thread memory region.

[0047] The isolated image metadata copy is a thread-private data object formed after access boundary locking and memory address mapping. It carries the surface radar backscattering coefficient, surface roughness parameters, spatial index, public parameter references, and abnormal state identifiers. This copy is the direct input object for S4 and S5. If a copy address conflict, public parameter reference failure, or copy inconsistency occurs during thread dispatching, the currently parallel task will first enter a waiting verification state. After the address is reconstructed, processing will continue, avoiding cross-thread read / write interference entering the scattering resolution stage.

[0048] S311: Create parallel environment parameters for independent cell replicas, construct a matching set of parallel execution tasks by combining parallel thread numerical indicators, allocate corresponding thread processing resources and memory resource quotas to the independent execution processes within the set of parallel execution tasks, and perform task packaging and allocation processing to obtain parallel execution task data.

[0049] Parallel environment parameters are data objects describing thread processing resources, memory resource quotas, task startup status, and replica binding relationships. A parallel execution task set is a collection of multiple parallel execution tasks, each corresponding to an independent cell replica and a thread memory address space. Task packaging and allocation processing combines replica identifiers, common parameter read references, thread resource states, and exception handling rules into parallel execution task data, enabling subsequent distribution phases to process tasks according to a unified task structure.

[0050] When creating data for parallel execution tasks, the integrity of the boundary records of independent cell replicas is first confirmed before binding the replicas to thread processing resources. If a replica's boundary record conflicts, the replica is not directly packaged; instead, a replica pending verification state is generated and returned to the space partitioning record. The task data that has been bound continues to enter S312 for memory address space distribution, ensuring that all tasks entering thread processing have a clear replica source and resource boundaries.

[0051] S312: Based on the parallel execution task data acquisition and allocation thread memory address space, the independent cell copy and preprocessing common parameters are distributed to the thread memory address space for storage management synchronization operation, establishing mutually independent parallel processing storage segment space and generating memory address distribution parameters.

[0052] The memory address distribution parameter is a data object that records the correspondence between independent cell copies, public parameter read references, and thread memory segments. Independent cell copies are stored as readable and writable objects after entering the thread's memory address space. Preprocessed public parameters are stored as read-only references or read-only copies when entering the thread's memory address space, enabling threads to read vegetation index data and incident angle conversion parameters at the corresponding spatial locations, but not to modify the original content of the public parameters.

[0053] Storage management synchronization operations are used to verify the consistency between replica data, public parameter references, and thread segment states. If the data item in the public parameters corresponding to the current replica space location is incomplete, a missing public parameter status is written to the thread's memory address space, allowing subsequent processing to skip this location or output an exception flag. If thread memory segment allocation fails, the task is executed in parallel to enter a reallocation state, and the reason for failure and the replica identifier are retained in the task record to ensure that the replica is not incorrectly allocated to other threads.

[0054] S313: Based on memory address distribution parameters, perform access boundary locking operations and out-of-bounds read / write interference shielding on independent cell replicas within the thread memory address space. Combined with access boundary locking operations, perform dedicated memory address mapping processing on memory segments to separate concurrent data items for each thread and establish isolated cell data replicas.

[0055] Access boundary locking restricts the range of accessible data based on the spatial boundaries of individual cell replicas, thread binding relationships, and memory segment records. Out-of-bounds read / write interference shielding detects whether a thread attempts to access an unbounded replica, a writable segment with common parameters, or another thread's memory segment. Once an out-of-bounds access request is detected, the current request is marked as a boundary exception and writing stops. The thread state record synchronously saves the exception type and target address type.

[0056] After the dedicated memory address mapping is completed, the isolated image metadata copy retains the surface radar backscattering coefficient, surface roughness parameters, spatial index, validity identifier, and common parameter reading path. This copy does not share a write area with other copies. Subsequently, S4 reads the surface roughness parameters and radar backscattering coefficient from this copy, and S5 reads the radar backscattering coefficient from the same copy and combines it with the vegetation layer attenuation to filter out vegetation scattering contributions, thus ensuring that the data source, processing boundary, and result attribution of the parallel processing process remain consistent.

[0057] Please see Figure 1 and Figure 5 S4: Extract surface roughness parameters and surface radar backscattering coefficients from isolated image data copies, map the radar microwave roughness surface scattering correlation between surface roughness parameters and incident angle conversion parameters, generate roughness surface scattering features, invert the underlying microwave reflection parameters between roughness surface scattering features and surface dielectric constant data, and obtain bare soil scattering cross section data.

[0058] Surface roughness parameters are data objects in isolated image data copies used to describe the impact of surface undulation on radar scattering. They originate from surface roughness inversion records or preceding surface parameter products associated with microwave radar remote sensing images. Surface radar backscattering coefficients are radar echo intensity representations of corresponding pixels in microwave radar remote sensing images, serving as the observational basis for scattering analysis in this step. Roughness surface scattering characteristics are intermediate feature objects formed by the radar microwave roughness surface scattering correlation between surface roughness parameters and incident angle transformation parameters, output to the bare soil scattering cross-section data generation process.

[0059] In this embodiment, the advanced integral equation model serves as the physical model for radar microwave roughness surface scattering mapping and underlying microwave reflection parameter inversion. This model receives surface roughness parameters, incident angle conversion parameters, and radar backscattering correlation data. As described in text, its processing logic involves separating the scattering contribution caused by the rough surface based on the scattering response relationship between radar observation geometry and surface roughness, and establishing an inverse correspondence between the roughness surface scattering characteristics and the surface dielectric constant data. In this embodiment, the surface dielectric constant data represents the category of physical parameters that influence microwave reflection behavior. These parameters originate from the medium parameter records required for microwave inversion processing or from prior parameter representations corresponding to the surface state, and are not presented in specific numerical form.

[0060] S411: Based on the isolated image data copy, perform parameter deconstruction, extract the surface radar backscattering coefficient and surface roughness parameters, perform matrix serialization and rearrangement of the extracted surface radar backscattering coefficient and surface roughness parameters, summarize the obtained data sequence and establish a radar microwave parameter set to obtain the surface microwave basic parameters.

[0061] Parameter deconstruction involves retrieving data items from isolated image metadata copies that are eligible for radar scattering analysis, based on the meaning of their data fields. Both the surface radar backscattering coefficient and surface roughness parameters have spatial indices and validity markers; they only proceed to matrix serialization and rearrangement when their spatial locations are consistent and their states are valid. Here, matrix serialization and rearrangement refers to organizing the data into a model-readable sequence according to spatial indices and field types, without involving formulaic expressions or numerical demonstrations.

[0062] The radar microwave parameter set is a data set consisting of surface radar backscattering coefficients, surface roughness parameters, spatial indices, validity identifiers, and common parameter reference relationships. The surface microwave fundamental parameters are the data objects output to S412 after processing this set. If backscattering data, roughness parameters, or spatial indices are found to be missing during the deconstruction process, an anomaly identifier for the corresponding location is generated. This identifier is passed forward along with the surface microwave fundamental parameters, and roughness surface scattering mapping is not subsequently performed at the anomaly locations.

[0063] S412: Extract the incident angle transformation parameters from the preprocessed common parameters, and perform radar microwave roughness surface scattering mapping based on the correlation between the surface roughness parameters and the incident angle transformation parameters using an advanced integral equation model, thereby generating roughness surface scattering features.

[0064] The incident angle transformation parameters are read from the preprocessed common parameters generated in S1 and matched according to the spatial index in the isolated image data copy. The processing object of radar microwave roughness surface scattering mapping is the correlation between surface roughness parameters and incident angle transformation parameters. The purpose of the processing is to identify the contribution of rough surfaces to the radar scattering response under corresponding observation geometry conditions. In this embodiment, roughness surface scattering characteristics specifically refer to the radar microwave roughness surface scattering correlation between surface roughness parameters and incident angle transformation parameters. Its data content includes spatial location, roughness correlation status, incident angle correlation status, and scattering feature identifiers that can be transferred to the inversion stage.

[0065] The advanced integral equation model is not output as a black box at this stage. Instead, features are extracted based on the physical mapping between the surface roughness and radar observation geometry. If the incident angle transformation parameter is missing in the common parameters, or if the surface roughness parameter has an anomaly in the isolated image data copy, no effective roughness surface scattering feature is generated at the corresponding location. Instead, a roughness surface scattering pending determination state is output. This state is then passed to S413 to determine whether the bare soil scattering cross-section data can form a valid result.

[0066] S413: The roughness surface scattering characteristics are imported into the advanced integral equation model. The underlying microwave reflection parameters between the roughness surface scattering characteristics and the surface dielectric constant are inversely retrieved. The microwave reflection parameter terms obtained from the inverse retrieval process are integrated to construct a microwave inversion cross-sectional data sequence, and the bare soil scattering cross-sectional data are output.

[0067] The inverse inversion process takes surface roughness scattering characteristics and surface dielectric constant data as input. It uses a high-level integral equation model to describe the radar scattering physics, distinguishing the scattering effects caused by rough surfaces from the underlying microwave reflection behavior. The underlying microwave reflection parameters are data items representing the microwave reflection response of bare soil surfaces. The microwave inversion cross-sectional data sequence is a data object formed by reorganizing the underlying microwave reflection parameters at each spatial location according to the image spatial index.

[0068] The bare soil scattering cross-section data is the direct input from S4 to S5, representing the bare soil scattering response used for subsequent vegetation attenuation analysis after removing the influence of roughness surface scattering relationships. If the roughness surface scattering characteristics are in a pending state or the surface dielectric constant data cannot be matched, a bare soil cross-section anomaly marker is written at the corresponding position in the bare soil scattering cross-section data, and the data is not entered into S5 as blank data. This anomaly marker is retained in the final result set stage to facilitate the distinction between inversion gaps caused by insufficient microwave reflection parameters and changes in inversion state caused by vegetation filtering.

[0069] Please see Figure 1 and Figure 6 S5: Generate vegetation radar transmittance parameters by mapping the vegetation index data and the incident angle transformation parameters between the vegetation canopy radar microwave transmittance features. Extract the vegetation canopy microwave attenuation ratio between the vegetation radar transmittance parameters and the bare soil scattering section data to generate the vegetation layer attenuation. Filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated pixel replicas to generate the parallel processing results of soil moisture content inversion.

[0070] The vegetation radar transmittance parameter is canopy microwave transmission characteristic data obtained by mapping vegetation index data and incident angle transformation parameters. It is used to express the influence of vegetation canopy on the radar microwave penetration process. Vegetation attenuation is a data object formed by the canopy attenuation relationship between the vegetation radar transmittance parameter and bare soil scattering cross-section data. It participates in the subsequent filtering of vegetation scattering contributions in the surface radar backscattering coefficient. The parallel processing result of soil moisture content inversion is output data collected by spatial indexing from the processing results of isolated image data copies from each thread. It includes valid inversion status, spatial location identifiers, and anomaly records.

[0071] In this embodiment, the water cloud model serves as a model for mapping the radar microwave transmittance characteristics of the vegetation canopy and generating vegetation attenuation. This model receives vegetation index data, incident angle conversion parameters, and bare soil scattering cross-section data. Through the physical relationship between vegetation cover status, radar observation geometry, and bare soil scattering response, it represents the transmission and attenuation effects of the vegetation canopy on radar signals. The output of the water cloud model is not presented as a formula or numerical example, but rather as a transmittance representation vector, attenuation status, and spatial index correspondence, which is then used in subsequent filtering processes.

[0072] S511: Extract vegetation index data and incident angle transformation parameters based on preprocessing common parameters, use the water cloud model to perform a positive mapping between vegetation index data and incident angle transformation parameters to construct a transmission vector and obtain vegetation radar transmission parameters.

[0073] The transmittance representation vector is a spatially indexed data object of vegetation canopy microwave transmittance characteristics, carrying the correspondence between vegetation index data, incident angle transformation parameters, canopy transmittance status, and validity indicators. The forward mapping process first reads vegetation index data and incident angle transformation parameters for the same spatial location from the preprocessed common parameters. Then, the water cloud model generates vegetation radar transmittance parameters based on the relationship between vegetation cover status and radar observation geometry. This parameter is output to S512 to establish a relationship between vegetation canopy microwave attenuation and bare soil scattering cross-section data.

[0074] If vegetation index data, incident angle conversion parameters, or spatially inconsistent states are missing from the common parameters, the water cloud model does not generate an effective transmittance representation vector for that spatial location. Instead, it outputs the canopy transmittance status to be determined. This status, along with the spatial index, is saved and used in S512 to determine whether the vegetation canopy microwave attenuation ratio can be extracted. Through this processing, the vegetation radar transmittance parameters remain traceable and will not be misused as effective canopy transmittance features due to missing common parameters.

[0075] S512: Input the vegetation radar transmittance parameters and bare soil scattering cross section data into the water cloud model, extract the vegetation canopy microwave attenuation ratio index between the vegetation radar transmittance parameters and bare soil scattering cross section data, summarize the vegetation canopy microwave attenuation ratio index and quantify the attenuation amplitude corresponding to each parallel processing node, and generate the vegetation layer attenuation amount.

[0076] In this embodiment, the vegetation canopy microwave attenuation ratio index is a textual engineering indicator used to describe the attenuation effect of canopy transmission characteristics on the scattering cross section of bare soil. It originates from the model mapping relationship between vegetation radar transmittance parameters and bare soil scattering cross section data. Quantifying the attenuation amplitude corresponding to each parallel processing node does not output specific values. Instead, it assigns the processing result of each thread to the attenuation state under the corresponding spatial index, forming a vegetation layer attenuation amount that can be used by S513 to filter out the vegetation scattering contribution.

[0077] During the extraction process, the water cloud model first verifies whether the vegetation radar transmittance parameters and bare soil scattering cross-section data originate from the same spatial location. Then, it generates an attenuation state based on the relationship between the vegetation canopy transmittance state and the bare soil scattering response. If the bare soil scattering cross-section data carries an anomaly marker, or if the vegetation radar transmittance parameters are in a pending state, the vegetation attenuation at the corresponding location is marked as unusable for filtering, and the source of the anomaly is retained in subsequent results. This processing ensures that the vegetation attenuation can both participate in filtering the scattering contribution of effective pixels and provide a clear feedback path for anomalous pixels.

[0078] S513: Filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated image data copy by using the vegetation layer attenuation amount, perform spatial variable inverse analysis on the residual data, and combine the surface radar backscattering cross section inversion analysis output terms corresponding to each parallel processing node to obtain the parallel processing result of soil moisture content inversion.

[0079] Vegetation scattering contribution filtering applies the vegetation layer attenuation to the surface radar backscattering coefficient in the isolated image data copy, separating the canopy scattering influence and retaining residual scattering information related to soil moisture status. The residual data, formed after vegetation scattering contribution filtering, carries a spatial index, residual backscattering status, bare soil scattering cross section reference, and anomaly status identifier. Spatial variable inverse analysis is used to map the residual data to the surface dielectric constant correlation status, bare soil scattering cross section data, and spatial index, forming the surface radar backscattering cross section inversion analysis output item for each parallel processing node.

[0080] When aggregating the output items corresponding to each parallel processing node, the thread results are backfilled into a unified result space according to the spatial index of the independent pixel replicas. If a thread result is missing, the backfill location conflicts, or the abnormal state is not confirmed, the source is first verified according to the replica boundary record, and the conflicting position is kept as an output state to be confirmed, without overwriting the existing valid results. The final parallel processing result of soil moisture content inversion includes both the inversion output of the available spatial location and the state records generated by missing image items, incomplete common parameters, failed scattering mapping, undetermined canopy transmittance, or abnormal thread boundaries, so that the subsequent reading process can distinguish between valid results and abnormal sources, and maintain the continuity of the multi-source remote sensing data parallel processing link.

Claims

1. A remote sensing data parallel processing method for soil moisture content inversion, characterized in that, Includes the following steps: S1: Acquire multispectral remote sensing images and microwave radar remote sensing images; extract vegetation indices based on band reflectance in the multispectral remote sensing images; map incident angle parameters according to observation data in the microwave radar remote sensing images; combine the vegetation indices with the incident angle parameters to generate preprocessing common parameters. S2: Obtain the total number of microwave image pixels and microwave image data storage capacity from the microwave radar remote sensing image, extract the cache capacity threshold of the local computing node, determine the number of parallel threads in combination with the number of available hardware cores, and perform spatial traversal and segmentation of the raster space corresponding to the total number of microwave image pixels according to the number of parallel threads to obtain independent pixel copies. S3: Construct a parallel execution task that matches the number of parallel threads, distribute the independent image data copy and the preprocessing common parameters to the thread memory address space allocated by the parallel execution task, perform access boundary locking and dedicated memory address mapping on the independent image data copy, and generate an isolated image data copy; S4: From the isolated image data copy, map the radar microwave roughness surface scattering correlation between the surface roughness parameters and the incident angle conversion parameters to generate roughness surface scattering features, and invert the underlying microwave reflection parameters between the roughness surface scattering features and the surface dielectric constant data to obtain bare soil scattering cross section data. S5: Map the vegetation canopy radar microwave transmittance features between the vegetation index data and the incident angle conversion parameters to generate vegetation radar transmittance parameters; extract the vegetation canopy microwave attenuation ratio between the vegetation radar transmittance parameters and the bare soil scattering cross section data to generate vegetation layer attenuation; filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated pixel replica; and generate parallel processing results for soil moisture content inversion.

2. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that: The reflectance of the red band and the reflectance of the near-infrared band of the multispectral pixels are extracted from the multispectral remote sensing image. The vegetation index data are generated by band feature fusion extraction based on the reflectance of the red band and the reflectance of the near-infrared band of the multispectral pixels. Satellite radar observation incident angle data is extracted from the microwave radar remote sensing image, and the units of the satellite radar observation incident angle data are unified from angle to radians to generate the incident angle conversion parameter.

3. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that: If the microwave image data storage capacity is greater than the cache capacity threshold, a first thread allocation instruction is generated based on the mapping relationship of twice the number of available hardware cores. If the microwave image data storage capacity is not greater than the cache capacity threshold, then a second thread allocation instruction is generated based on the equal number of available hardware cores. The target concurrency level is determined based on the first thread allocation instruction or the second thread allocation instruction, and the number of parallel threads is generated.

4. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that: The roughness surface scattering characteristics specifically refer to the radar microwave roughness surface scattering correlation between the surface roughness parameters and the incident angle conversion parameters.

5. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that, The specific steps of S1 are as follows: S111: Acquire multispectral remote sensing images and microwave radar remote sensing images, extract multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance from the multispectral remote sensing images, perform band feature fusion extraction on multispectral pixel red band reflectance and multispectral pixel near-infrared band reflectance, and generate vegetation index data. S112: Based on microwave radar remote sensing imagery, the observation data is mapped to extract satellite radar observation incident angle data. The satellite radar observation incident angle data is then converted to radians. The processed data is then converted and processed in conjunction with the microwave radar remote sensing imagery observation data mapping relationship to obtain the incident angle conversion parameters. S113: Perform sequence combination and splicing processing based on the vegetation index data and the incident angle conversion parameters. Perform spatial correspondence set combination on the corresponding dimension data columns of the vegetation index data and the incident angle conversion parameters. List multi-dimensional parameters according to the combined data items and extract the corresponding observation data items to construct a unified format architecture and establish preprocessing common parameters.

6. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that, The specific steps of S2 are as follows: S211: Acquire microwave radar remote sensing images, extract the total number of microwave image pixels and microwave image data storage capacity, collect the number of available hardware cores of local computing nodes, simultaneously extract the local computing node cache capacity threshold, summarize various hardware-related parameters and image storage parameter characteristics, and generate node hardware basic characteristic data. S212: Based on the microwave image data storage capacity in the node hardware basic feature data, compare it with the cache capacity threshold. If the microwave image data storage capacity is greater than the cache capacity threshold, perform a double core quantity relationship mapping in combination with the number of available hardware cores. If it is not greater, perform an equal core quantity relationship mapping to obtain thread allocation instruction parameters. S213: Based on the thread allocation instruction parameters, determine the target number of concurrent operations and convert it into a parallel thread numerical index. According to the converted parallel thread numerical index, perform spatial traversal and segmentation on the grid space corresponding to the total number of microwave image pixels, extract the set of associated pixels of each grid image, and establish an independent pixel copy.

7. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that, The specific steps of S3 are as follows: S311: Create parallel environment parameters for the independent cell replica, construct a matching parallel execution task set in combination with the parallel thread numerical index, allocate corresponding thread computing resources and memory resource quotas to each independent execution unit within the parallel execution task set, and perform task packaging and allocation processing to obtain parallel execution task data. S312: Based on the parallel execution task data, obtain and allocate the thread memory address space, distribute the independent cell copy and the previously mentioned preprocessing common parameters to the thread memory address space for storage management and synchronization operations, establish mutually independent parallel computing storage segment space, and generate memory address distribution parameters; S313: Based on the memory address distribution parameters, perform access boundary locking operations and out-of-bounds read / write interference shielding on the independent cell replicas within the thread memory address space. Combined with the access boundary locking operation, perform dedicated memory address mapping operations on memory segments to separate concurrent data items of each thread and establish isolated cell data replicas.

8. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that, The specific steps of S4 are as follows: S411: Based on the isolated image data copy, perform parameter deconstruction, extract the surface radar backscattering coefficient and surface roughness parameters, perform matrix serialization and rearrangement of the extracted surface radar backscattering coefficient and surface roughness parameters, summarize the obtained data sequence and establish a radar microwave parameter set to obtain the surface microwave basic parameters. S412: Extract the incident angle conversion parameter from the preprocessed common parameters, and perform radar microwave roughness surface scattering mapping on the correlation between the surface roughness parameter and the incident angle conversion parameter in the surface microwave basic parameters using an advanced integral equation model to generate roughness surface scattering features. S413: The roughness surface scattering characteristics are imported into the advanced integral equation model. The underlying microwave reflection parameters between the roughness surface scattering characteristics and the surface dielectric constant are inversely inverted. The microwave reflection parameter terms obtained from the inverse inversion operation are integrated to construct a microwave inversion cross-sectional data sequence, and the bare soil scattering cross-sectional data are output.

9. The parallel processing method for remote sensing data for soil moisture content retrieval according to claim 1, characterized in that, The specific steps of S5 are as follows: S511: Extract vegetation index data and incident angle conversion parameters according to the preprocessing common parameters, use the water cloud model to perform a positive mapping of the vegetation canopy radar microwave transmittance characteristics between the vegetation index data and the incident angle conversion parameters, construct a transmittance representation vector, and obtain the vegetation radar transmittance parameters. S512: Input the vegetation radar transmittance parameters and the bare soil scattering cross section data into the water cloud model, extract the vegetation canopy microwave attenuation ratio index between the vegetation radar transmittance parameters and the bare soil scattering cross section data, summarize the vegetation canopy microwave attenuation ratio index and quantify the attenuation amplitude corresponding to each calculation node, and generate the vegetation layer attenuation amount. S513: Filter out the vegetation scattering contribution of the surface radar backscattering coefficient in the isolated image data copy using the vegetation layer attenuation amount, perform spatial variable reverse analysis calculation on the residual data, and aggregate the surface radar backscattering cross section inversion analysis output items corresponding to each parallel computing node to obtain the parallel processing result of soil moisture content inversion.

Images